diff --git a/internal/application/service/session.go b/internal/application/service/session.go
index 26bf9ff4..f8c8643a 100644
--- a/internal/application/service/session.go
+++ b/internal/application/service/session.go
@@ -342,15 +342,12 @@ func (s *sessionService) GenerateTitleAsync(
 	modelID string,
 	eventBus *event.EventBus,
 ) {
-	// Extract values from context before cloning
-	tenantID := ctx.Value(types.TenantIDContextKey)
+	// Use session's tenant for title generation (session belongs to session.TenantID; ctx may have effectiveTenantID when using shared agent)
+	sessionTenantID := session.TenantID
 	requestID := ctx.Value(types.RequestIDContextKey)
 	go func() {
-		// Create new background context and copy values
-		bgCtx := context.Background()
-		if tenantID != nil {
-			bgCtx = context.WithValue(bgCtx, types.TenantIDContextKey, tenantID)
-		}
+		// Create new background context with session owner's tenant so ListModels/GetChatModel/Update use correct scope
+		bgCtx := context.WithValue(context.Background(), types.TenantIDContextKey, sessionTenantID)
 		if requestID != nil {
 			bgCtx = context.WithValue(bgCtx, types.RequestIDContextKey, requestID)
 		}
diff --git a/internal/handler/session/helpers.go b/internal/handler/session/helpers.go
index d0cd6614..81752abf 100644
--- a/internal/handler/session/helpers.go
+++ b/internal/handler/session/helpers.go
@@ -160,6 +160,7 @@ func (h *Handler) setupStreamHandler(
 func (h *Handler) setupStopEventHandler(
 	eventBus *event.EventBus,
 	sessionID string,
+	sessionTenantID uint64,
 	assistantMessage *types.Message,
 	cancel context.CancelFunc,
 ) {
@@ -167,7 +168,9 @@ func (h *Handler) setupStopEventHandler(
 		logger.Infof(ctx, "Received stop event, cancelling async operations for session: %s", sessionID)
 		cancel()
 		assistantMessage.Content = "用户停止了本次对话"
-		h.completeAssistantMessage(ctx, assistantMessage)
+		// Use session's tenant for message update (ctx may have effectiveTenantID when using shared agent)
+		updateCtx := context.WithValue(ctx, types.TenantIDContextKey, sessionTenantID)
+		h.completeAssistantMessage(updateCtx, assistantMessage)
 		return nil
 	})
 }
diff --git a/internal/handler/session/qa.go b/internal/handler/session/qa.go
index 71385cdd..651101b6 100644
--- a/internal/handler/session/qa.go
+++ b/internal/handler/session/qa.go
@@ -210,7 +210,7 @@ func (h *Handler) setupSSEStream(reqCtx *qaRequestContext, generateTitle bool) *
 	}
 
 	// Setup stop event handler
-	h.setupStopEventHandler(eventBus, reqCtx.sessionID, reqCtx.assistantMessage, cancel)
+	h.setupStopEventHandler(eventBus, reqCtx.sessionID, reqCtx.session.TenantID, reqCtx.assistantMessage, cancel)
 
 	// Setup stream handler
 	h.setupStreamHandler(asyncCtx, reqCtx.sessionID, reqCtx.assistantMessage.ID,
@@ -413,7 +413,9 @@ func (h *Handler) executeNormalModeQA(reqCtx *qaRequestContext, generateTitle bo
 
 			logger.Infof(streamCtx.asyncCtx, "Knowledge QA service completed for session: %s", sessionID)
 			// Content already contains <think>...</think> tags from chat_completion_stream.go
-			h.completeAssistantMessage(streamCtx.asyncCtx, streamCtx.assistantMessage)
+			// Use session's tenant for message update (asyncCtx may have effectiveTenantID when using shared agent)
+			updateCtx := context.WithValue(streamCtx.asyncCtx, types.TenantIDContextKey, reqCtx.session.TenantID)
+			h.completeAssistantMessage(updateCtx, streamCtx.assistantMessage)
 			// Emit EventAgentComplete - this will trigger handleComplete which sends the SSE complete event
 			// Note: Don't cancel context here, let the SSE handler close naturally after receiving the complete event
 			streamCtx.eventBus.Emit(streamCtx.asyncCtx, event.Event{
@@ -517,7 +519,9 @@ func (h *Handler) executeAgentModeQA(reqCtx *qaRequestContext) {
 					errors.NewInternalServerError(fmt.Sprintf("Agent QA service panicked: %v\n%s", r, string(buf))),
 					map[string]interface{}{"session_id": sessionID})
 			}
-			h.completeAssistantMessage(streamCtx.asyncCtx, streamCtx.assistantMessage)
+			// Use session's tenant for message update (session belongs to session.TenantID; asyncCtx may have effectiveTenantID when using shared agent)
+			updateCtx := context.WithValue(streamCtx.asyncCtx, types.TenantIDContextKey, reqCtx.session.TenantID)
+			h.completeAssistantMessage(updateCtx, streamCtx.assistantMessage)
 			logger.Infof(streamCtx.asyncCtx, "Agent QA service completed for session: %s", sessionID)
 		}()